Scholarlink

<p>The curing and shaping of composite laminates primarily rely on the autoclave molding process in the aerospace industry. However, the non-uniform temperature field in the autoclave may lead to non-isothermal curing, which affects the molding accuracy. To address it, this paper proposes a framework that combines the finite element method with machine learning models, in which Bayesian Optimization (BO), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Light Gradient Boosting Machine (LightGBM) are adopted to construct the BO-CNN-LSTM and BO-CNN-LightGBM models for effectively predicting the temperature histories at representative monitoring points of composite laminates during curing. First, a multi‑field coupled finite element simulation model was established, and experiments on the temperature field in the autoclave were conducted to validate the accuracy of the finite element model. Temperature curves at different measuring points on the laminate under various process parameters were obtained through the finite element model, and then these data were used to train the machine learning models for prediction.This study is positioned as a monitoring-point-oriented prediction of curing-temperature histories, aiming to support efficient thermal-response evaluation of composite laminates under different autoclave process conditions. The results show that the heating ramp and the dwell step during the autoclave molding significantly influence the laminate’s temperature histories at the monitoring points. The finite element model can accurately reflect the variation trends in temperature curves over process parameters. Both machine learning models demonstrated good predictive capabilities in predicting temperature curves. But BO‑CNN‑LSTM outperforms BO‑CNN‑LightGBM in capturing the complex dynamic relationship between the representative-point temperature histories and process parameters, with a prediction error rate below 5.7%, thereby providing a rapid basis for evaluating temperature-history responses at representative monitoring points.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid FEM and Machine Learning Method for Predicting Temperature Histories at Representative Monitoring Points During Composite Laminate Curing

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Jingyu Zhang,
  • Xiaobo Yang,
  • He Xiang,
  • Lihua Zhan,
  • Xintong Wu

摘要

The curing and shaping of composite laminates primarily rely on the autoclave molding process in the aerospace industry. However, the non-uniform temperature field in the autoclave may lead to non-isothermal curing, which affects the molding accuracy. To address it, this paper proposes a framework that combines the finite element method with machine learning models, in which Bayesian Optimization (BO), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Light Gradient Boosting Machine (LightGBM) are adopted to construct the BO-CNN-LSTM and BO-CNN-LightGBM models for effectively predicting the temperature histories at representative monitoring points of composite laminates during curing. First, a multi‑field coupled finite element simulation model was established, and experiments on the temperature field in the autoclave were conducted to validate the accuracy of the finite element model. Temperature curves at different measuring points on the laminate under various process parameters were obtained through the finite element model, and then these data were used to train the machine learning models for prediction.This study is positioned as a monitoring-point-oriented prediction of curing-temperature histories, aiming to support efficient thermal-response evaluation of composite laminates under different autoclave process conditions. The results show that the heating ramp and the dwell step during the autoclave molding significantly influence the laminate’s temperature histories at the monitoring points. The finite element model can accurately reflect the variation trends in temperature curves over process parameters. Both machine learning models demonstrated good predictive capabilities in predicting temperature curves. But BO‑CNN‑LSTM outperforms BO‑CNN‑LightGBM in capturing the complex dynamic relationship between the representative-point temperature histories and process parameters, with a prediction error rate below 5.7%, thereby providing a rapid basis for evaluating temperature-history responses at representative monitoring points.